Understanding thermal comfort and wellbeing of older South Australians using occupant-centric models

Abstract

The proportion of older people (i.e., those aged 65 or over) in the world’s population is increasing due to historically low fertility rates combined with increased life expectancy. In order to respond to these demographic trends, a growing body of policy and research over the last decades has accepted that ageing-in-place is most beneficial in the interests of older people’s independence, health and wellbeing, as well as to reduce the economic burden on governments and society for the provision of aged care facilities. While there are several guidelines that provide information about designing dwellings to suit ageing-in-place, information to aid older people’s thermal comfort and related wellbeing is not always considered. This thesis addresses the current knowledge on thermal comfort of older people in order to provide environments that meet their individual requirements and help improve their overall wellbeing. Traditionally, thermal comfort standards adopt aggregate modelling approaches as the bases on which to establish the requirements for human occupancy in the built environment. Aggregate models explain thermal comfort at a population level, which can result in limitations in real scenarios as individual thermal perceptions can vary significantly. In recent years, a growing number of studies have been conducted to address these limitations by developing ‘personal comfort models’. Instead of an average response from a large population, personalised models predict individuals’ thermal comfort by using a single person’s direct feedback. Nonetheless, up until the research presented in this thesis, studies on personal comfort models have focused on younger adults, generally in office environments. This presents a critical research gap because intergroup heterogeneity in personal capabilities and needs tends to be greater among older people, causing the use of aggregate models for older adults to result in even more frequent exposure to unacceptable thermal environments. These, in turn, can interact with multiple comorbidities, leading to adverse health outcomes and possibly premature institutional care. Thus, personalising models hold the promise of a more accurate way to predict older people’s thermal comfort and to manage their thermal environments better. Considering the issues and opportunity presented above, the goal of this research is to advance the current knowledge on the use of personal thermal comfort models by focusing on older people in their home environments. The research aims to achieve this goal by: (1) reviewing the present understandings of personal comfort models, (2) investigating older people’s’ thermal environment, behaviours and preferences; (3) developing personal comfort models for older people and comparing the results with the predictions by established aggregate models; and (4) investigating the application of personal comfort models in managing the thermal environment of older people. Two indoor environmental monitoring field studies and related point-in-time thermal comfort surveys were conducted to collect datasets for the analyses. The first dataset was collected from 71 older adults in 57 households located in South Australia across 9 months. This was followed by the application of deep learning (i.e., a class of machine learning) to develop personal comfort models for 28 out of these 71 participants using different combinations of the collected series of indoor environmental measurements, along with behavioural and health/wellbeing survey answers. The second dataset was collected during shorter 2-week periods involving 11 of the original 71 participants, during which, in addition to measuring the indoor environmental parameters and collecting behavioural and health/wellbeing survey answers, the participants’ hand skin temperatures were measured. The development of personal models for 4 of these participants was then conducted, including skin temperatures as an additional modelling input. Several performance indicators, including average accuracy, Cohen’s Kappa Coefficient and Area Under the Receiver Operating Characteristic Curve (AUC) were employed to assess the skill of the developed individual models. All models’ performance indicators were then compared with a ‘version’ of the Predicted Mean Vote (PMV) model, termed, in this thesis, the PMVc. The results showed that the 28 personal thermal comfort models for older adults that used environmental, behavioural and health/wellbeing perception as input variables presented an average accuracy of 74%, an average Cohen’s Kappa Coefficient of 61% and an average (AUC) of 0.83. This represented a significant improvement in predictive performance when compared with the generalised PMVc model, which presented an average accuracy of 50%, an average Cohen’s Kappa Coefficient of 24%, and an average AUC of 0.62. Similarly, the exploration with the 4 personal comfort models adding skin temperature measurements to the abovementioned input variables, and excluding health/wellbeing perception − which yielded slightly lower performance when included −, resulted in an average accuracy of 67%, an average Cohen’s Kappa Coefficient of 50% and an average AUC of 0.77. This also represented a superior predictive performance of the individualised models when compared with the PMVc model. In order to investigate the applications of the personal comfort models in operation, two participants were selected as case studies and their respective personal models were tested for their ability to estimate personal heating and cooling temperature set points, using calibrated building performance simulation models. The simulated energy loads derived from the use of personal set points were compared with simulated energy loads using 21°C as the heating set point and 24°C as the cooling set point, which represented the common averaged set points used in building simulation studies. The results show that, using the personal set points, good agreement between the actual and simulated heating and cooling energy loads was achieved. When comparing the error ratios with the ones resulting from simulations assuming a 21°C set point for heating and a 24°C for cooling, the study also showed that the personal set points significantly outperformed these traditional assumptions. Finally, as a secondary application exploration, one selected participant’s personal model was converted to a smart phone Application (App) format to examine the opportunity to use the model as a web-based smart phone tool to aid designers and caregivers to manage the thermal environments of older people by considering individual requirements. This conversion also proved to be successful, allowing the automatic calculation of thermal preference for the selected participant, thereby demonstrating its potential to aid designers and caregivers. The novelty and therefore the contributions of this research lay in different areas. Whilst the literature on personal comfort models has focussed solely on younger adults in office environments, this research has explored a methodology for predicting thermal comfort of older people in their dwellings. Additionally, it has introduced health/wellbeing perception as a predictor of thermal preference – a variable often overlooked in architectural sciences and building engineering. Finally, the research indicates that, compared with aggregated models, personal models provide superior utility in predicting an individual’s preferred thermal environment, which therefore offers the potential for more accurate tools to design and improve older people’s living environments so that wellbeing is optimised, healthy ageing is fostered and autonomy while ageing is prolonged. The research recommends a range of topics for future investigation, such as the models’ misclassification costs and the integration among wearable sensors, predictors and actuators in the context of older people. In addition, the development of standard protocols necessary for the models’ deployment in real scenarios is prescribed. In conclusion, the research demonstrates that, as a concept, personal comfort models have the ability to absorb people’s diversity in the context of their environmental conditions, and may therefore represent an important step towards providing knowledge aimed at enhancing wellbeing and improving the overall resilience of the built environment.